基于卷积神经网络的压印字符预处理

Embossed characters preprocessing based on convolutional neural network

压印字符被广泛用于制造中以记录产品信息,高效地识别压印字符具有很大的价值。与传统的光学字符识别(OCR)不同,图像增强是识别压印字符的重要步骤。文中提出了一种基于卷积神经网络(CNN)的增强算法,以改善压印字符与背景之间的对比度。然后,结合研究对象即鞋底本身的特点,提出了一种预处理方法将鞋底圆孔去除:利用CNN和图像像素值相与计算法结合去除鞋底圆孔,之后,可以将常规的OCR算法直接应用于处理后的图像。对带有压印字符的鞋底进行了实验,结果表明该算法可以有效地对压印字符进行预处理。

Embossed characters are widely used in manufacturing to record product information,and identifying embossed characters efficiently has great value.Unlike traditional optical character recognition(OCR),image enhancement is an important step in recognizing embossed characters.An enhancement algorithm is proposed based on convolutional neural networks(CNN)to improve the contrast between embossed characters and the background.Then,combined with the characteristics of the research object-the sole itself,a pre-processing method is proposed to remove the round hole of the sole:a combination of CNN and image pixel value and calculation method is used to remove the round hole of the sole,After that,the conventional OCR algorithm can be directly applied to the processed image.Experiments are carried out on the soles with embossed characters.The results show that the algorithm can effectively preprocess the embossed characters.